A few enhancements for fv3jedi to avoid extra post processing of the analysis

[TingLei-NOAA]

In the current regional fv3jedi cycling, multiple post processing of the analysis (from variational and enkf) are needed, which include : creation of updated D wind ; creation of updated delp ; creation from scratch of the analysis files ( and hence, non-analysis variables in the restart files needed to be added in the state variables or be taken care of in the post-processing step). This issue is to document currently explored enhancements of fv3jedi, mainly within fv3jedi fortran io interface to allow fvj3jedi cycling without need to those postprocessing while the new codes still abide to the fv3jedi design/conventions.

1. For D wind problem, this draft PR had been shown not abiding by the fv3jedi design
   https://github.com/JCSDA-internal/fv3-jedi/pull/1472
2. For D wind problem, an new implementation
   https://github.com/JCSDA-internal/fv3-jedi/pull/1473
3. In-place writing capability for regional restart
   An initial version of in-place capability for regional fv3jedi write restart file  TingLei-daprediction/sandbox-fv3-jedi#4

[SamuelDegelia-NOAA]

@TingLei-NOAA Thanks a ton for working on all of these optimizations. All of these will be great to use in rrfs-workflow. Let me know when (2,3) are ready and I can test them on our GETKF test case for na3km where I/O is the biggest bottleneck.

[TingLei-NOAA]

@SamuelDegelia-NOAA Thanks. Yes. those draft PR had passed my sanity tests ( successfully finished, and the values looked ok).
That will be great if you would test and evaluate them in your cases.

[SamuelDegelia-NOAA]

Sounds good - I'll start some tests!

[TingLei-NOAA]

A new draft PR on Ting 's fork : TingLei-daprediction/sandbox-fv3-jedi#7.
On top of the parallel regional restart and write inplace, A increment would be converted to D wind increment and added to D background. Please see some recent discussions at https://github.com/JCSDA-internal/fv3-jedi/pull/1473